AI Agents in 2025 - Where to Start & What Really Works (No Hype)



AI Summary

Summary of Video Transcript: The Year of AI Agents with Joe Mora

Introduction

  • The episode discusses the significance of AI agents in the current year.
  • Joe Mora, CEO and founder of Crew AI, is the guest.
  • The focus is on understanding what an agent is and how to build agents for current and future roles.

What is an Agent?

  • Agents are different from simple predictive tools like LLMs (e.g., GPT and Tropic) because they have agency.
  • Agents have tools and memory, which allow them to perform tasks autonomously.
  • The discussion includes the difference between short-term and long-term memory in agents.

The Year of Agents

  • OpenAI launched Operator as part of Chat GPT Pro Edition, which is a browser control agent.
  • Operator can perform tasks like making reservations and doing research, but it is currently slow.
  • Joe Mora shares his experience with agents helping him post on social media, emphasizing their value despite being slower than humans.

Adoption of Agents

  • Agents are being executed in the millions every month, indicating their growing use.
  • Enterprises are adopting agents quickly, with impacts on their bottom lines.
  • The adoption is often led by a central department like CIO or CTO, ensuring control and integration with company systems.

UX Patterns for Agents

  • There’s a debate on the right UX pattern for agents, whether users should see agents work or trust them to work autonomously.
  • Joe Mora discusses the importance of being able to inspect what agents are doing for security and reassurance.

Building and Deploying Agents

  • Companies are advised to start with low precision, low-risk use cases for agents and then scale up.
  • There are no-code platforms for less technical people and frameworks like CI for more technical users.
  • Agents are being adopted for various roles within companies, but it’s still early days with only a small percentage of features in production.

Use Cases for Agents

  • Agents are being used for tasks like enriching customer data, creating sales hypotheses, and automating A/B testing for SEO.
  • The agents can autonomously perform tasks based on simple inputs from humans.

Expectations vs. Reality

  • There is a mismatch between expectations and the current state of agent technology.
  • Agents are not yet ready to take over entire departments or work without human intervention.
  • The adoption of agents is still in the early stages, and there are technical challenges to overcome.

Conclusion

  • Joe Mora emphasizes the potential of agents and suggests starting with simple use cases.
  • He recommends using platforms like Crew AI to explore and build agents.
  • The episode concludes with a discussion on the future of employment, where agents could play a significant role in an individual’s job performance.

Detailed Instructions and URLs

  • No specific CLI commands, website URLs, or detailed instructions were provided in the transcript.